Parameter efficeint fine tuning LLMs and faster inference